Author(s):
Need: Chemistry is an important gateway course for STEM careers, and the quantitative problem solving aspects of this course are one of the biggest barriers to student success. This project is developing the ORCCA (Open Response Chemistry Cognitive Assistant) tutors to support students as they learn to solve problems in introductory chemistry courses. A central goal is to allow students to solve problems from a blank slate, just as they would with paper and pencil, and to provide scaffolding only when necessary. Current online problem-solving tools either provide feedback only on the final result (minimal scaffolding) or use a step-by-step interface that does not provide practice with how to develop problem-solving strategies (heavy scaffolding). By coupling a user interface for free-form entry of mathematical expressions with a production rule system that monitors student work, ORCCA aims to provide scaffolding only when needed (dynamic scaffolding). By developing production rules that cover most of the quantitative reasoning in chemistry, we also aim to support a wide variety of problem types. Guiding Question: To what extent can an artificial intelligence (AI) system, based on a single set of production rules, follow student work on a wide range of problems and provide hints and feedback that improve student learning?Outcomes: The ORCCA system was developed based on the Cognitive Tutor Authoring Tools previously created at Carnegie Mellon. We have shown that a single set of production rules are able to support problem solving on a broad range of problems, including stoichiometry, thermochemistry, chemical equilibrium, and acid-base chemistry. To date over 1000 students have used the ORCCA system in General Chemistry courses. The User Interface has been refined through pilot testing and retroactive think-aloud interview research. This work has led to an optimized user experience where roughly 90% of the time a student who asks for a hint at any stage of the problem being solved receives one that is relevant to the step in the process that is being completed.Broader Impacts: Because problem solving is central to all STEM fields, the information being gained from investigating how AI methods involving production rules for chemistry problems can add to the understanding of building more flexible tutoring programs in most areas of science. In addition, the nature of chemistry problem solving poses challenges for AI methods as well, so this project holds the promise of enhancing development of AI technology for education as well.
Coauthors
Emily Smith, Iowa State University, Ames, IA; David Yaron, Maxwell Benson, Sandra Raysor, Jonathon Sewall, Ken Koedinger, Vincent Aleven, Carnegie Mellon University, Pittsburgh, PA
